Semantic approach to text entailment for question answering - new domain for uncertainty modeling

In this work we present semantic based methodologies to improve the accuracy of question answering (QA) systems. In particular, we concentrate on the textual entailment module, which is implemented as one of the major modules of our QA system and is used to rank order the retrieved passages/sentences from the search engine. Using semantic parsing methods, we extract grammatical relation and semantic structure of sentences in corpus as well as the question posed by the user. The textual entailment module is used to match the retrieved sentences and the question using this semantic information. We also use different ontology to extract the terms from the sentences and the question to enrich the search query results. The entailment module reveals the possibility of the entailment between the retrieved sentences that are highly likely to contain the answer phrase and the question and rank the sentences accordingly. We introduce the uncertainty modeling methods we have been building for the QA text entailment module to improve the accuracy of the system.